Re: AW: SVN v1.6.5 (CollabNet) and SVN v1.7.0 (WANdisco) creates many threads, allocate all free memory and crashes/stops

Здравствуйте, Markus.

Вы писали 18 октября 2011 г., 17:55:55:

>> We had used SVN v1.6.5 server (CollabNet build) on Windows 2003 x64 with
>> 4GB RAM.
>>
>> Yesterday SVN server is stopped unexpectedly.
>>
>> I tried to restart it, but behaviour was:
>> immediately after start it created about 800-900 threads, and allocate
>> about 800-900mb of memory than stops (or crashes; but seems, stops with
>> error that there no resources -- I can's say exactly for all cases, just I
>> saw this error later with 1.7.0 in my debugging experiments).

MS> If svnserve 1.6.5 worked fine for a long time, I would think that
MS> this is not caused by a bug in 1.6.5, but some different reason.

I think it is a bug in server code, that was in 1.6.5, and that exists
in 1.7.0. Just a "floating" bug, that have a difficult steps to
reproduce (or some external environment required).

MS> Maybe defective RAM,

No. There is no any problems. For Windows NT-based systems this
usually case BSOD, not a usermode problems, moreover for one program
only.

MS> maybe some malware infection on the server,

No. Just we a information security company. ;)
There is no internet access, and all other access is really stricted
to this server.

MS> maybe a corrupted repository (defective Hard-Disk?),

It was a first idea, that because of low memory available at some
moment, repository (or cache db) may be corrupted in some strange way.
But if this idea is correct, than bug must occur EVERY start. But
currently server working without any problems with the same
repository!

MS> or maybe some (D)DOS attack or amok running script.
No, there is no other connection, it is an internal server.

MS> Some ideas helping to hunt down the bug: (If you have enough
MS> resources to do that?)

MS> Did you try to replace the repository with the latest backup?

No, but I thought about this, when upgrade to 1.7.0 did not helped.
But because bug is not reproducable more, we do nothing now.
Just working and awaiting when it will appear again. I really want to
stop in debugger and to see what code is doing this allocations/create
threads.

MS> Could you try to use the latest stable 1.6.X svnserve?

We used 1.6.5 for a few years when this problem appears.

MS> Could you try to use a dump/restore or svnsync copy of the repository?

This is idea, if the problem appear again and it will not be possible
to ignore it. But anyway, I think that even if DB have some problems,
in this case it is a bug in code that reading DB, because sometimes
the problem appears, and some time - not.

MS> If you use BDB backend, you could even try to dump/restore to FSFS, or vice versa.

We use FSFS.

Thank you.

-- 
С уважением,
 Andrey                          mailto:andrey_at_online-solutions.ru